CN115856204A - Method for predicting gas concentration of tunneling working face based on three-dimensional echo state network - Google Patents
Method for predicting gas concentration of tunneling working face based on three-dimensional echo state network Download PDFInfo
- Publication number
- CN115856204A CN115856204A CN202211085414.XA CN202211085414A CN115856204A CN 115856204 A CN115856204 A CN 115856204A CN 202211085414 A CN202211085414 A CN 202211085414A CN 115856204 A CN115856204 A CN 115856204A
- Authority
- CN
- China
- Prior art keywords
- data
- gas concentration
- state network
- echo state
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 68
- 230000005641 tunneling Effects 0.000 title claims abstract description 44
- 238000012549 training Methods 0.000 claims abstract description 39
- 238000012545 processing Methods 0.000 claims abstract description 29
- 239000003245 coal Substances 0.000 claims abstract description 16
- 230000002159 abnormal effect Effects 0.000 claims abstract description 14
- 238000007781 pre-processing Methods 0.000 claims abstract description 14
- 230000008569 process Effects 0.000 claims abstract description 10
- 238000000354 decomposition reaction Methods 0.000 claims description 32
- 238000004364 calculation method Methods 0.000 claims description 12
- 238000012935 Averaging Methods 0.000 claims description 10
- 238000012360 testing method Methods 0.000 claims description 9
- 238000011156 evaluation Methods 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 6
- 238000003672 processing method Methods 0.000 claims description 4
- 238000001228 spectrum Methods 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 2
- 238000012216 screening Methods 0.000 claims description 2
- 230000009286 beneficial effect Effects 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 6
- 238000004880 explosion Methods 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 210000002569 neuron Anatomy 0.000 description 3
- 230000009467 reduction Effects 0.000 description 3
- 238000009825 accumulation Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 210000004556 brain Anatomy 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 210000000225 synapse Anatomy 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to the technical field of coal mine safety, in particular to a method for predicting gas concentration of a tunneling working face based on a three-dimensional echo state network, which comprises the following steps: s1, acquiring gas concentration data of a coal mine driving face in a mine in real time, preprocessing the acquired gas data, and processing data complexity to obtain a driving face gas concentration data set for training an echo state network. The training method of the three-dimensional echo state network based on the CEEMDAN-SE can effectively process random and large amount of data, generate a network learning result by distributing different weights, effectively reduce time cost, improve operation efficiency, and have considerable robustness to the result in the training process, thereby being beneficial to improving the accuracy of prediction and identification of gas data, accurately searching abnormal data characteristics in the predicted data, and intervening safety measures.
Description
Technical Field
The invention relates to the technical field of coal mine safety, in particular to a method for predicting gas concentration of a tunneling working face based on a three-dimensional echo state network.
Background
The gas concentration over-accumulation is a root cause of gas explosion accidents, is an extremely complex and dangerous dynamic disaster, controls the gas accumulation, mainly enhances the gas management, enhances the gas monitoring and control, and adheres to a gas inspection system, finds that the gas is over-limit and is reported in time, the gas concentration causing the gas explosion is in a range, and the flame can not be automatically spread when the gas mixture with the concentration lower than the lower explosion limit or higher than the upper explosion limit contacts with an ignition source. Therefore, how to predict and identify the occurrence of the phenomenon is particularly important, the method is based on a deep learning algorithm, abnormal data characteristics are searched in predicted data by analyzing the gas concentration in real time and training an echo state network model, and abnormal prediction monitoring data are transmitted to workers in an alarm form to intervene safety measures. The traditional gas concentration data prediction method comprises a support vector machine, a decision tree, a traditional regression method and a time sequence decomposition algorithm of a statistical learning method.
At present, aiming at the characteristics of time-varying property, instability and nonlinearity of gas concentration data and combining with the reason that the gas data volume is large and the complexity is high, the traditional prediction method cannot adapt to the characteristics of the data, the characteristics of the data are difficult to identify and process, and an ideal result is difficult to obtain. The echo state network in the deep learning algorithm is just suitable for processing data with large data volume and complexity, a large number of nodes and neurons in the echo state network can be used for training the data, and the characteristics of the data are memorized, so that the echo state network has certain advantages in data training and prediction.
However, when the first data is obtained, the data is still in a high imbalance state, the capacity of the existing echo state network for preprocessing the data is limited, the existing echo state network is mostly a one-dimensional echo state network, and after the data transmitted by the underground sensor is processed in one dimension, the accuracy of prediction and identification of the gas data still needs to be improved, so that the accuracy of searching abnormal data characteristics in the predicted data is not facilitated, and safety measure intervention is performed.
Disclosure of Invention
The invention aims to solve the defects in the prior art and provides a tunneling working face gas concentration prediction method based on a three-dimensional echo state network.
In order to achieve the above purposes, the technical scheme adopted by the invention is as follows: the method for predicting the gas concentration of the tunneling working surface based on the three-dimensional echo state network comprises the following steps:
s1, acquiring gas concentration data of a driving face under a mine in real time, preprocessing the acquired gas data, and processing data complexity to obtain a gas concentration data set of the driving face, which can be used for training an echo state network, required by a user;
s2, learning gas concentration data of a tunneling working face by adopting a three-dimensional echo state network of CEEMDAN-SE, and then performing data training to obtain a prediction result;
s3, measuring a plurality of prediction models through different measuring indexes, comparing the correlation and loss between the true value and the predicted value of the models, and selecting the model with the minimum error and the minimum loss;
s4, deploying the optimal model into a safety control server of the coal mine, and inputting the latest captured gas concentration data into the model; if the calculated result is not in the safe and controllable numerical range, the gas safety environment of people can be judged to be abnormal and needs to be processed in time, otherwise, the condition is normal;
preferably, the method for acquiring and processing the working face gas concentration data in S1 includes the following steps:
s101, collecting data of gas sensors of all working faces in real time from a coal mine gas disaster wind direction control platform;
s102, performing modal decomposition on gas concentration data obtained from different sensor positions according to CEEMDAN preprocessing experience;
s103, carrying out sample entropy calculation data complexity classification on the data set subjected to modal decomposition processing;
and S104, circularly traversing the whole gas concentration data to obtain the same sequence numbers co-IMF1, co-IMF2 and co-IMF3 which are from the same sequence data but have three different types of complexity.
Preferably, the data preprocessing empirical mode decomposition method in S102 includes the following steps:
s102-1, reading data;
s102-2, setting the processing times K of the original data;
s102-3, respectively adding random white noise to the K original data to form a series of new data;
s102-4, performing EMD decomposition on the series of new data to obtain a series of IMF components;
s102-5, respectively averaging IMF components of corresponding modes to obtain EEMD decomposition results;
s102-6, carrying out ensemble averaging on the modal components after EEMD decomposition, carrying out ensemble averaging calculation on the CEEMDAN decomposition after the obtained first-order IMF component to obtain a final first-order IMF component, and then repeating the operation on the residual part.
Preferably, the processing method for classifying the complexity of the data set in S103 includes the following steps:
s103-1, carrying out sample entropy calculation on the collected K IMF component sequence data obtained by CEEMDAN decomposition;
s103-2, calculating the sequence complexity of each IMF component through sample entropy, and dividing the K IMF components into three types according to the complexity range of different sequence data, namely a high-frequency eigenmode component co-IMF1, a medium-frequency modal component co-IMF2 and a low-frequency modal component co-IMF3.
Preferably, the data processing method of the three-dimensional echo state network of CEEMDAN-SE in S2 includes the following steps:
s201, segmenting watt concentration co-IMF1, co-IMF2 and co-IMF3 into a training set D-co-IMF1, D-co-IMF2 and D-co-IMF3 and a testing set T-co-IMF1, T-co-IMF2 and T-co-IMF3;
s201, selecting a framework of an echo state network model, and then improving the echo state network into a three-dimensional echo state network capable of inputting three-dimensional data (in the model training process, determining alpha as a leakage rate, and optimizing the network by using an Adam optimizer);
s203, determining the number of nodes, the spectrum radius SR, the size N of the reserve pool, the input unit size IS of the reserve pool, the sparsity degree SD of the reserve pool and the regularization factor, wherein each change of the parameters generates a new model, and the experimental result IS closely related to the parameters, so that the fitting or overfitting of the experimental result due to incorrect selection of the parameters IS avoided.
Preferably, the gas concentration data set co-IMF in S201 is divided into a training set D-co-IMF and a prediction set T-co-IMF according to the proportion of 7.
Preferably, in S202, a framework of the echo state network model is selected, and then the one-dimensional echo state network is improved into a three-dimensional echo state network capable of inputting three-dimensional data;
the improved three-dimensional echo state network formula is as follows:
given three segments of signals:
u(0),u(1),…,u(Nt-1)；
l(0),l(1),…,l(Nt-1)；
w(0),w(1),…,w(Nt-1)；
target value:
v(1),v(2),…,v(Nt)；
m(1),m(2),…,m(Nt)；
y(1),y(2),…,y(Nt)；
predicting the value:
v(Nt+1),v(Nt+2),...；
m(Nt+1),m(Nt+2),...；
y(Nt+1),y(Nt+2),...；
wherein:
u∈R M*3 ，W IR ∈R N*M ，W res ∈R N*N ，r∈R N*3 ，W RO ∈R L*N ,v∈R L*3 ；W IR ∈R N*M ,W res ∈R N*N are all given a predetermined numberValue, only need to calculate W in the calculation process RO ∈R L*N And (4) finishing.
Operation from input to library:
W IR *u(t)，W IR *l(t)，W IR *w(t)；
update of r (t) in the library:
r(t+Δt)＝f[W res *r(t)+W IR *u(t)]，
r(t+Δt)＝f[W res *r(t)+W IR *l(t)]，
r(t+Δt)＝f[W res *r(t)+W IR *w(t)]；
from library to output:
u(t+Δt)＝W RO *r(t+Δt)，
l(t+Δt)＝W RO *r(t+Δt)，
w(t+Δt)＝W RO *r(t+Δt)；
loss function:
preferably, the method for comparing the model in the S3 comprises the following steps:
s301, inputting the test co-IMF into the model obtained in S2 respectively to obtain model prediction co-IMF';
s302, calculating the evaluation index values of MAE and RMSE between the real test values and the predicted values of different models;
s303, screening the model with the minimum loss of the MAE and the RMSE as an optimal model.
The MAE formula is calculated as follows:
the formula for calculating RMSE is:
wherein
Is a model predicted value>
Is a real prediction value.
Preferably, in order to ensure timeliness and high accuracy of the model, model training needs to be performed periodically, and an optimal model is selected.
Preferably, the prediction system related to the gas concentration prediction method comprises the following module units:
the working face gas concentration data acquisition unit is used for acquiring gas concentration data of a coal mine driving working face in a mine in real time;
the working face gas concentration data processing unit is used for preprocessing the obtained gas data and processing the data complexity to obtain a tunneling working face gas concentration data set which can be used for training an echo state network;
the working face gas concentration data tunneling unit is used for learning and tunneling the working face gas concentration data by adopting a three-dimensional echo state network of CEEMDAN-SE, and then performing data training to obtain a prediction result;
the prediction model comparison unit is used for measuring a plurality of prediction models through different indexes, comparing the correlation and the loss between the true value and the predicted value of the models and selecting the model with the minimum error and the minimum loss;
and the latest captured gas concentration data processing and judging unit is used for deploying the optimal model into a safety control server of the coal mine, inputting the latest captured gas concentration data into the model, and if the calculated result is not in a safe and controllable numerical range, judging that the gas safety environment is abnormal and needs to be processed in time, otherwise, judging that the condition is normal.
Compared with the prior art, the invention has the following beneficial effects:
the invention provides a method for predicting gas concentration of a tunneling working face based on a three-dimensional echo state network, which can be used for processing data modal decomposition and calculation complexity and amplifying correlation and modal property among data; the training method of the three-dimensional echo state network based on the CEEMDAN-SE can effectively process random and large amount of data, generate a network learning result by distributing different weights, effectively reduce time cost, improve operation efficiency, and have considerable robustness on the result during training, thereby being beneficial to improving the accuracy of prediction and identification of gas data, and accurately searching abnormal data characteristics in the predicted data to intervene safety measures.
Drawings
In the present invention, in order to more clearly describe the embodiments of the present invention, the following briefly describes the implementation steps in the form of the attached drawings.
Fig. 1 is a schematic flow diagram of a method for predicting gas concentration of a tunneling working face based on a three-dimensional echo state network, provided by the invention;
FIG. 2 is a gas concentration data intelligent system prediction frame diagram of the gas concentration prediction method of the tunneling working face based on the three-dimensional echo state network;
fig. 3 is an empirical mode decomposition diagram of a self-adaptive noise complete set for data preprocessing of the method for predicting the gas concentration of the tunneling working surface based on the three-dimensional echo state network provided by the invention;
FIG. 4 is a prediction effect diagram of the method for predicting the gas concentration of the tunneling working face based on the three-dimensional echo state network, provided by the invention;
FIG. 5 is a prediction performance index table of each prediction model comparing gas data of a tunneling working face according to the method for predicting gas concentration of a tunneling working face based on a three-dimensional echo state network provided by the invention;
fig. 6 is a system block diagram of the method for predicting the gas concentration of the tunneling working surface based on the three-dimensional echo state network.
In the figure: 1. a working face gas concentration data acquisition unit; 2. a working face gas concentration data processing unit; 3. a working face gas concentration data tunneling unit; 4. a prediction model comparison unit; 5. and a processing and judging unit for newly captured gas concentration data.
Detailed Description
The following description is presented to disclose the invention so as to enable any person skilled in the art to practice the invention. The preferred embodiments in the following description are given by way of example only, and other obvious variations will occur to those skilled in the art.
In order to better understand the method for predicting gas concentration in the present invention, the terms related to the embodiments of the present invention will be explained first.
CEEMDAN algorithm
The CEEMDAN method achieves a reconstruction error of almost 0 with a small number of averaging times by adding adaptive white noise at each decomposition stage.
EMD is proposed by Huang equal to 1998, and the method can adaptively decompose Intrinsic Mode Functions (IMFs) with different frequencies for an original signal and is strong in adaptability. EEMD is an improved method based on EMD, and mainly adds different white Gaussian noises into an original sequence for multiple times, then EMD decomposition is respectively carried out, and finally the obtained IMF components are averaged to obtain a final result, so that the occurrence of modal aliasing is avoided. However, the white noise added by the EEMD method is not completely cancelled after multiple averaging. The algorithm depends on the magnitude and the average number of white noise added. The CEEMDAN method achieves a reconstruction error of almost 0 with a small number of averaging times by adding adaptive white noise at each decomposition stage. Therefore, the CEEMDAN method can overcome the modal aliasing phenomenon existing in EMD, and also solves the problems of incompleteness of EEMD decomposition and low calculation efficiency caused by the fact that reconstruction errors must be reduced by increasing the average times.
SE algorithm
Sample Entropy (SE) is a new method proposed by Richman in 2000 and capable of measuring time sequence complexity, and is an improvement on Approximate Entropy (AE), so that dependency on time sequence length is reduced, and errors of approximate entropy in a calculation process can be effectively reduced. Given a PM10 historical concentration value sequence { x (i) |1 ≦ i ≦ N }, m is a mode dimension, and r is a similarity tolerance.
Three-dimensional echo state network
As a novel neural network Echo State Network (ESN), the network is a network which is based on the basic principle of a neural network in biology, and simulates a neural system of a human brain to a complex information processing mechanism on the basis of network topology knowledge after understanding and abstracting a human brain structure and an external stimulus response mechanism. The echo state network is composed of an input layer, a reserve pool and an output layer. The neurons in the reservoir are interconnected to retain the information left at the previous time. The connection weights from the input layer of the echo state network to the reserve pool and in the reserve pool are all generated by random initialization. In the training process, only the connection weight from the reserve pool to the output layer needs to be trained, which becomes a linear regression problem, so that the echo state network is trained very fast. The method can overcome some defects of the traditional neural network, such as low training efficiency, low convergence speed of the algorithm and the like, and is also commonly used for solving practical problems. However, in the research of the echo state network, there are many unsolved problems, for example, the full connection of output synapses leads to the reduction of the network prediction performance, the random connection of neurosynaptic in the reserve pool leads to the reduction of the network prediction precision, and the random connection between neurons in the reserve pool leads to the reduction of the network prediction performance.
Improved three-dimensional echo state network formula
Given three segments of signals:
u(0),u(1),…,u(Nt-1)；
l(0),l(1),…,l(Nt-1)；
w(0),w(1),…,w(Nt-1)；
and a target value:
v(1),v(2),…,v(Nt)；
m(1),m(2),…,m(Nt)；
y(1),y(2),…,y(Nt)；
predicting the value:
v(Nt+1),v(Nt+2),...；
m(Nt+1),m(Nt+2),...；
y(Nt+1),y(Nt+2),...；
wherein:
u∈R M*3 ，W IR ∈R N*M ，W res ∈R N*N ，r∈R N*3 ，W RO ∈R L*N ,v∈RL*3；W IR ∈R N*M ,W res ∈R N*N are all given values in advance, and only W needs to be calculated in the calculation process RO ∈R L*N And (4) finishing.
Operation from input to library:
W IR *u(t)，W IR *l(t)，W IR *w(t)；
update of r (t) in the library:
r(t+Δt)＝f[W res *r(t)+W IR *u(t)]，
r(t+Δt)＝f[W res *r(t)+W IR *l(t)]，
r(t+Δt)＝f[W res *r(t)+W IR *w(t)]；
from library to output:
u(t+Δt)＝W RO *r(t+Δt)，
l(t+Δt)＝W RO *r(t+Δt)，
w(t+Δt)＝W RO *r(t+Δt)；
loss function:
secondly, the method for predicting the gas concentration of the tunneling working face based on the three-dimensional echo state network is implemented specifically:
referring to fig. 1 to 6, a method for predicting gas concentration of a tunneling working face based on a three-dimensional echo state network is provided, which comprises the following steps:
in the gas safety system, processing original safety information, extracting historical gas concentration data from a database of a mine control center, preprocessing and analyzing the historical gas concentration data, and then establishing an available historical tunneling working face gas concentration prediction data set;
dividing the obtained available historical tunneling working face gas concentration prediction data set into two types, wherein one type is a training set, the other type is a testing set, training the constructed CEEMDAN-SE three-dimensional echo state network through the training set, and performing comparison testing on the trained model through the testing set;
inputting real-time gas concentration data into the trained echo state network model, and obtaining a prediction result of the model of the invention as a prediction value of the gas concentration;
wherein processing the original security information comprises:
s1, acquiring underground gas concentration data in real time, performing data preprocessing on the acquired historical gas concentration data, and performing empirical mode decomposition to obtain a gas data set which can be used for training an echo state network in the invention;
detecting each index data in the frame in real time through a sensor installed in a mine, transmitting the data detected in real time to an on-line manner, and acquiring underground gas concentration data in real time;
s2, learning gas concentration data of a tunneling working face by adopting a three-dimensional echo state network of CEEMDAN-SE, and then performing data training to obtain a prediction result;
for a gas data set obtained underground, processing the data by a self-adaptive noise complete set empirical mode decomposition (CEEMDAN) method, and setting E i (. Cndot.) is the i-th eigenmode component obtained by EMD decomposition and the i-th eigenmode component obtained by CEEMDAN decompositionEigenmode component
u j J =1,2, \ 8230for a gaussian white noise signal satisfying a standard normal distribution, n. the number of times white noise is added, epsilon is a standard table of white noise addition, and y (t) is a signal to be decomposed. The CEEMDAN decomposition steps are as follows:
reading data to be decomposed;
adding Gaussian white noise to the number y (t) to be decomposed to obtain a new signal:
y(t)+(-1) q εv j (t)
q =1,2, EMD decomposition is carried out on the new data to obtain a first-order eigenmode component C 1 . The formula is as follows:
the first eigenmode component of the CEEMDAN decomposition is obtained by ensemble averaging the N generated mode components:
calculating the residual error after removing the first modal component:
at r 1 (t) adding positive and negative paired Gaussian white noise to obtain new data, and performing EMD decomposition by using the new data as a carrier to obtain a first-order modal component D 1 From this, the 2 nd eigenmode component of the CEEMDAN decomposition can be obtained:
calculating the residual error after the second modal component is removed:
and repeating the steps, knowing that the obtained residual error data is a monotonous function, and ending the algorithm, wherein the decomposition cannot be continued. When the number of eigenmode components obtained at this time is K, the raw data y (t) is decomposed into:
and evaluating the complexity of a time sequence corresponding to the K intrinsic mode components according to sample entropy (SampEn), and dividing the K intrinsic mode components into three types, namely a high-frequency intrinsic mode component co-IMF1, a medium-frequency mode component co-IMF2 and a low-frequency mode component co-IMF3 according to different complexities.
Inputting the obtained co-IMF1, co-IMF2 and co-IMF3 into a three-dimensional echo state network for training, wherein the size of a library needs to be determined in the echo state network, the spectrum radius SR =1.5, the size of a reserve pool N =2000, the average degree D =2, the regularization constant eta =1e-4 and the random seed =20000, and the network activation function is tanh.
And obtaining training prediction result sequences co-IMF ', co-IMF ' and co-IMF ', and reconstructing and adding the results to obtain the final required prediction result.
And S3, measuring a plurality of different models through different measuring indexes, and comparing the correlation of the true value and the predicted value of the models to select the model with the minimum error and the minimum loss.
Respectively calculating the average absolute error value and the root mean square error value of a plurality of models, wherein the model corresponding to the two parameter values which are as small as possible is selected;
and S4, applying the optimal model to the field of gas safety for predicting the gas concentration of the tunneling working face, if the calculated result is not in a safe and controllable numerical value interval, judging that the gas safety environment is abnormal, and processing in time, otherwise, judging that the condition is normal.
In order to ensure that the model can obtain a more accurate result, the model needs to be updated frequently, and network data training needs to be performed again at intervals so as to select the optimal model, namely, the data of the past four months are trained for the past 36 days, in the example, the training set time length is four months;
the prediction system related to the gas concentration prediction method comprises the following module units:
the working face gas concentration data acquisition unit 1 is used for acquiring gas concentration data of a coal mine driving working face in a mine in real time;
the working face gas concentration data processing unit 2 is used for preprocessing the obtained gas data and processing the data complexity to obtain a tunneling working face gas concentration data set which can be used for training an echo state network;
the working face gas concentration data tunneling unit 3 is used for learning and tunneling working face gas concentration data by adopting a three-dimensional echo state network of CEEMDAN-SE, and obtaining a plurality of models according to different model structures and different training methods;
the prediction model comparison unit 4 is used for measuring a plurality of prediction models through different indexes, comparing the correlation and loss between the true value and the predicted value of the models, and selecting the model with the minimum error and the minimum loss;
and the latest captured gas concentration data processing and judging unit 5 is used for deploying the optimal model to a safety control server of the coal mine, inputting the latest captured gas concentration data into the model, judging that the gas safety environment is abnormal if the calculated result is not in a safe and controllable numerical value interval, and processing in time if the calculated result is not in the safe and controllable numerical value interval, otherwise, judging that the gas safety environment is normal.
Finally, the results are demonstrated through experiments
The data used in the experiment are collected from an intelligent accurate early warning system for the five-tiger mountain gas disasters, and comprise data of a gas workbench from 2021 year to 2022 year, wherein the data in the period from 2021-11-01 to 2022-02-28 is collected, 1 data is read every few minutes, and 25920 data are collected as a gas concentration data set.
FIG. 4 is a prediction effect diagram of a gas concentration prediction method for a tunneling working face based on a three-dimensional echo state network, and it can be seen that the prediction result has the characteristics of high prediction accuracy, small loss and high robustness. Through multiple predictions, the experimental model combines the prediction set of the coal mine tunneling working face gas concentration prediction method of the self-adaptive noise complete set empirical mode decomposition-sample entropy (CEEMDAN-SE) and the three-dimensional echo state network (3D-ESN) and the evaluation index RMSE of the training set is 1.311, and the evaluation index MAE is 1.142. The evaluation index RMSE of the same data set and the traditional ESN in a prediction set and a training set is 4.151, the evaluation index MAE is 3.543, and the results of multiple experiments tend to be stable and are higher than the evaluation index of the prediction set and the training set of the coal mine tunneling working face gas concentration prediction method combining self-adaptive noise complete set empirical mode decomposition-sample entropy (CEEMDAN-SE) and a three-dimensional echo state network (3D-ESN);
therefore, the model of the invention can better track the fluctuation trend of the gas concentration time series, and has the minimum prediction evaluation index value and the highest prediction precision. The method has the characteristics of practicability and accuracy in the application of the method in the field of gas safety, so that the accuracy of prediction and identification of gas data is improved, abnormal data characteristics can be accurately searched in the predicted data, abnormal prediction monitoring data are transmitted to workers in an alarm mode, and safety measure intervention is carried out.
The foregoing shows and describes the general principles, essential features, and advantages of the invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are merely illustrative of the principles of the invention, but that various changes and modifications may be made without departing from the spirit and scope of the invention, which is defined by the appended claims and their equivalents.
Claims (10)
1. The method for predicting the gas concentration of the tunneling working surface based on the three-dimensional echo state network is characterized by comprising the following steps of:
s1, acquiring gas concentration data of a coal mine driving face in a mine in real time, preprocessing the acquired gas data, and processing data complexity to obtain a driving face gas concentration data set for training an echo state network;
s2, learning gas concentration data of a tunneling working face by adopting a three-dimensional echo state network of CEEMDAN-SE, and then performing data training to obtain a prediction result;
s3, measuring the relevant prediction models through different indexes, comparing the relevance and loss between the true value and the predicted value of the models, and selecting the model with the minimum error and the minimum loss;
and S4, deploying the optimal model into a safety control server of the coal mine, inputting the latest captured gas concentration data into the model, judging whether the gas safety environment is abnormal and needs to be processed in time if the calculated result is not in a safety controllable numerical value interval, and otherwise, judging that the condition is normal.
2. The method for predicting the gas concentration of the tunneling working surface based on the three-dimensional echo state network according to claim 1, wherein the method for acquiring and processing the gas concentration data of the working surface in the step S1 comprises the following steps:
s101, collecting data of gas sensors of all working faces in real time from a coal mine gas disaster wind direction control platform;
s102, carrying out modal decomposition on gas concentration data obtained from different sensor positions according to CEEMDAN preprocessing experience;
s103, carrying out sample entropy calculation data complexity classification on the data set subjected to modal decomposition processing;
and S104, circularly traversing the whole gas concentration data to obtain the same sequence numbers co-IMF1, co-IMF2 and co-IMF3 which are from the same sequence data but have three different complexity ranges.
3. A method for predicting gas concentration of a tunneling working surface based on a three-dimensional echo state network according to claim 2, wherein the empirical mode decomposition method for data preprocessing in S102 comprises the following steps:
s102-1, reading data;
s102-2, setting the processing times K of the original data;
s102-3, respectively adding random white noise to the K original data to form a series of new data;
s102-4, performing EMD decomposition on the series of new data to obtain a series of IMF components;
s102-5, respectively averaging IMF components of corresponding modes to obtain EEMD decomposition results;
s102-6, carrying out ensemble averaging on the modal components after EEMD decomposition, carrying out ensemble averaging calculation on the CEEMDAN decomposition after the obtained first-order IMF component to obtain a final first-order IMF component, and then repeating the operation on the residual part.
4. The method for predicting the gas concentration of the tunneling working face based on the three-dimensional echo state network according to claim 3, wherein the processing method for classifying the complexity of the data set in S103 comprises the following steps:
s103-1, carrying out sample entropy calculation on the collected K IMF component sequence data obtained by CEEMDAN decomposition;
s103-2, calculating the sequence complexity of each IMF component through sample entropy, and then dividing the K IMF components into three types according to the complexity of different sequence data, namely a high-frequency eigenmode component co-IMF1, a medium-frequency modal component co-IMF2 and a low-frequency modal component co-IMF3.
5. The method for predicting the gas concentration of the tunneling working face based on the three-dimensional echo state network according to claim 4, wherein the data processing method of the CEEMDAN-SE three-dimensional echo state network in S2 comprises the following steps:
s201, segmenting watt concentration co-IMF1, co-IMF2 and co-IMF3 into a training set D-co-IMF1, D-co-IMF2 and D-co-IMF3 and a testing set T-co-IMF1, T-co-IMF2 and T-co-IMF3;
s202, selecting a framework of a one-dimensional echo state network model, and then improving the one-dimensional echo state network into a three-dimensional echo state network capable of inputting three-dimensional data;
s203, determining the number of nodes, the spectrum radius SR, the size N of the reserve pool, the input unit size IS of the reserve pool, the sparsity degree SD of the reserve pool and the regularization factor, wherein each change of the parameters generates a new model, and the experimental result IS closely related to the parameters, so that the fitting or overfitting of the experimental result due to incorrect selection of the parameters IS avoided.
6. The method for predicting the gas concentration of the tunneling working face based on the three-dimensional echo state network according to claim 4, wherein the gas concentration data co-IMF in the S201 is divided into a training set D-co-IMF and a prediction set T-co-IMF according to a proportion of 7.
7. The method for predicting the gas concentration of the tunneling working face based on the three-dimensional echo state network according to claim 5, wherein the step S202 of improving the data processing formula of the three-dimensional echo state network comprises the following steps:
the improved three-dimensional echo state network formula is as follows:
given three segments of signals:
u(0),u(1),...,u(Nt-1)；
l(0),l(1),…,l(Nt-1)；
w(0),w(1),…,w(Nt-1)；
target value:
v(1),v(2),...,v(Nt)；
m(1),m(2),...,m(Nt)；
y(1),y(2),...,y(Nt)；
predicting the value:
v(Nt+1),v(Nt+2),...；
m(Nt+1),m(Nt+2),...；
y(Nt+1),y(Nt+2),...；
wherein:
u∈R M*3 ，W IR ∈R N*M ，W res ∈R N*N ，r∈R N*3 ，W RO ∈R L*N ，v∈R L*3 ，W IR ∈R N*M ,W res ∈R N*N are all given values in advance, and only W needs to be calculated in the calculation process RO ∈R L*N Then the method is finished;
operation from input to library:
W IR *u(t)，W IR *l(t)，W IR *w(t)；
update of r (t) in the library:
r(t+Δt)＝f[W res *r(t)+W IR *u(t)]，
r(t+Δt)＝f[W res *r(t)+W IR *l(t)]，
r(t+Δt)＝f[W res *r(t)+W IR *w(t)]；
from library to output:
u(t+Δt)＝W RO *r(t+Δt)，
l(t+Δt)＝W RO *r(t+Δt)，
w(t+Δt)＝W RO *r(t+Δt)；
loss function:
8. a tunneling working face gas concentration prediction method based on a three-dimensional echo state network according to claim 7, characterized in that the model comparison method in S3 comprises the following steps:
s301, respectively inputting the test set D-co-IMF into the model obtained in S2, and obtaining a model prediction result which is co-IMF;
and S302, calculating the evaluation index values of the MAE and the RMSE between the real test values and the predicted values of the different models.
S303, screening the model with the minimum loss of the MAE and the RMSE as an optimal model.
9. The method for predicting the gas concentration of the tunneling working surface based on the three-dimensional echo state network according to claim 8, wherein in order to ensure the timeliness and high accuracy of the model, model training is periodically carried out to select the optimal model.
10. A gas concentration prediction method for a heading face based on a three-dimensional echo state network according to claim 9, wherein the prediction system related to the gas concentration prediction method comprises the following module units:
the working face gas concentration data acquisition unit (1) is used for acquiring gas concentration data of a coal mine driving working face in a mine in real time;
the working face gas concentration data processing unit (2) is used for preprocessing the obtained gas data and processing the data complexity to obtain a tunneling working face gas concentration data set which can be used for training an echo state network;
the working face gas concentration data tunneling unit (3) is used for learning and tunneling the working face gas concentration data by adopting a three-dimensional echo state network of CEEMDAN-SE, and then performing data training to obtain a prediction result;
the prediction model comparison unit (4) is used for measuring the correlation prediction model through different indexes, comparing the correlation and loss between the true value and the predicted value of the model, and selecting the model with the minimum error and the minimum loss;
and the latest captured gas concentration data processing and judging unit (5) is used for deploying the optimal model into a safety control server of the coal mine, inputting the latest captured gas concentration data into the model, judging that the gas safety environment is abnormal if the calculated result is not in a safe and controllable numerical range, and processing the abnormal gas safety environment in time, otherwise, judging that the condition is normal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211085414.XA CN115856204B (en) | 2022-09-06 | 2022-09-06 | Tunneling working face gas concentration prediction method based on three-dimensional echo state network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211085414.XA CN115856204B (en) | 2022-09-06 | 2022-09-06 | Tunneling working face gas concentration prediction method based on three-dimensional echo state network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115856204A true CN115856204A (en) | 2023-03-28 |
| CN115856204B CN115856204B (en) | 2023-06-23 |
Family
ID=85660756
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211085414.XA Active CN115856204B (en) | 2022-09-06 | 2022-09-06 | Tunneling working face gas concentration prediction method based on three-dimensional echo state network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115856204B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117454762A (en) * | 2023-10-30 | 2024-01-26 | 昆明理工大学 | Coal-penetrating tunnel face gas concentration prediction method of Markov-neural network |
| CN117828419A (en) * | 2023-12-05 | 2024-04-05 | 中国矿业大学 | Mine gas concentration prediction method based on optimization variation modal decomposition |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106503792A (en) * | 2016-10-25 | 2017-03-15 | 辽宁工程技术大学 | A kind of gas density Forecasting Methodology based on adaptive mode massing neutral net |
| CN108921359A (en) * | 2018-07-26 | 2018-11-30 | 安徽大学 | A kind of distribution gas density prediction technique and device |
| CN110334451A (en) * | 2019-07-09 | 2019-10-15 | 精英数智科技股份有限公司 | Determine the method, apparatus and system of the factor and trend that influence variation ofgas density |
| CN113887637A (en) * | 2021-10-09 | 2022-01-04 | 安徽理工大学 | Coal and gas outburst accident prediction and identification method based on deep cycle neural network |
| US20220099610A1 (en) * | 2020-06-17 | 2022-03-31 | Nexceris, Llc | Systems and methods for monitoring a gas analyte |
-
2022
- 2022-09-06 CN CN202211085414.XA patent/CN115856204B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106503792A (en) * | 2016-10-25 | 2017-03-15 | 辽宁工程技术大学 | A kind of gas density Forecasting Methodology based on adaptive mode massing neutral net |
| CN108921359A (en) * | 2018-07-26 | 2018-11-30 | 安徽大学 | A kind of distribution gas density prediction technique and device |
| CN110334451A (en) * | 2019-07-09 | 2019-10-15 | 精英数智科技股份有限公司 | Determine the method, apparatus and system of the factor and trend that influence variation ofgas density |
| US20220099610A1 (en) * | 2020-06-17 | 2022-03-31 | Nexceris, Llc | Systems and methods for monitoring a gas analyte |
| CN113887637A (en) * | 2021-10-09 | 2022-01-04 | 安徽理工大学 | Coal and gas outburst accident prediction and identification method based on deep cycle neural network |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117454762A (en) * | 2023-10-30 | 2024-01-26 | 昆明理工大学 | Coal-penetrating tunnel face gas concentration prediction method of Markov-neural network |
| CN117454762B (en) * | 2023-10-30 | 2024-05-24 | 昆明理工大学 | Coal-penetrating tunnel face gas concentration prediction method of Markov-neural network |
| CN117828419A (en) * | 2023-12-05 | 2024-04-05 | 中国矿业大学 | Mine gas concentration prediction method based on optimization variation modal decomposition |
| CN117828419B (en) * | 2023-12-05 | 2024-06-07 | 中国矿业大学 | Mine gas concentration prediction method based on optimization variation modal decomposition |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115856204B (en) | 2023-06-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109657547B (en) | Accompanying model-based abnormal trajectory analysis method | |
| CN109492830B (en) | Mobile pollution source emission concentration prediction method based on time-space deep learning | |
| Silva et al. | A novel unsupervised approach based on a genetic algorithm for structural damage detection in bridges | |
| CN115856204B (en) | Tunneling working face gas concentration prediction method based on three-dimensional echo state network | |
| CN112819207B (en) | Geological disaster space prediction method, system and storage medium based on similarity measurement | |
| CN115828140B (en) | Method, system and application for detecting fault by fusing neighborhood mutual information and random forest | |
| CN108241901B (en) | Transformer early warning evaluation method and device based on prediction data | |
| CN111179591B (en) | Road network traffic time sequence characteristic data quality diagnosis and restoration method | |
| CN116341901B (en) | Integrated evaluation method for landslide surface domain-monomer hazard early warning | |
| CN113901977A (en) | Deep learning-based power consumer electricity stealing identification method and system | |
| CN113988210A (en) | Method and device for restoring distorted data of structure monitoring sensor network and storage medium | |
| CN116205544A (en) | Non-invasive load identification system based on deep neural network and transfer learning | |
| CN111967308A (en) | Online road surface unevenness identification method and system | |
| CN117972314B (en) | Cloud platform monitoring method and system based on digital twinning | |
| CN110866558A (en) | Multi-source data fusion analysis-based rotating equipment state early warning method | |
| CN110533109A (en) | A kind of storage spraying production monitoring data and characteristic analysis method and its device | |
| CN113742983A (en) | Long-span structural damage identification method based on depth self-encoder neural network | |
| CN111126694A (en) | Time series data prediction method, system, medium and device | |
| CN116186507A (en) | Feature subset selection method, device and storage medium | |
| CN113688506A (en) | Potential atmospheric pollution source identification method based on multidimensional data such as micro-station | |
| KR20210086096A (en) | Method and apparatus for detecting detailed unusual behavior in satellite | |
| Merrill et al. | Deep convolution for irregularly sampled temporal point clouds | |
| Hong et al. | Drainage network flow anomaly classification based on XGBoost | |
| CN117726241B (en) | Building structure quality detection method, equipment and storage medium | |
| Khelil et al. | Soft Sensing Modeling Based on Support Vector Regression and Self-Organizing Maps Model Selection for Water Quality Monitoring |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |